' $ A Concepts-Focused Introduction to Functional Programming Using Standard ML Section 5— Each-Of And One-Of Types (Records and Datatypes) Draft of August 12, 2010 &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Where are we $ • Done features: functions, tuples, lists, local bindings, options • Done concepts: syntax vs. semantics, environments, mutation-free • Today features: records, datatypes, case expressions (pattern-matching) • Today concepts: “One-of” types, constructors/extractors, case-coverage &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Base types and compound types $ Languages typically provide a small number of “built-in” types and ways to build compound types out of simpler ones: • Base types examples: int, string, unit • Type builder examples: tuples, lists, records (see code) Base types clutter a language definition; better to make them libraries when possible. • ML does this to a remarkable extent (e.g., we will soon define away bool and conditionals) &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Compound-type flavors $ Conceptually, just a few ways to build compound types: 1. “Each-of”: A t contains a t1 and a t2 2. “One-of”: A t contains a t1 or a t2 3. “Self-reference”: The definition of t may refer to t Examples: • int * bool (syntactic sugar for a record type in ML) • int option • int list Remarkable: A lot of data can be described this way. (optional jargon: product types, sum types, recursive types) &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'User-defined types $ There are many reasons to define your own types: 1. Using a tuple with 12 fields is incomprehensible 2. Writing down large types is unpleasant; we have computers for that 3. Large programs can use abstract types to be robust to change • See later section 4. So the language doesn’t have to “bake in” lists and options and ... &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Datatype $ One-of types are less similar across languages • We’ll discuss OO’s approach to one-of later In ML, we make a new type with a datatype binding, e.g.: datatype mytype = TwoInts of int*int | Str of string | Pizza Semantics: Extend the environment with three constructors (in part, functions/constants that produce values of type mytype) • TwoInts has type int*int->mytype • Str has type string->mytype • Pizza has type mytype. So we have a way to build them... what’s missing? &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'The old way $ For lists and options, we had a way to: • Test which variant a value was (e.g., null) • Extract the values from value-carrying variants (e.g., hd, tl) – Makes no sense if you have the wrong variant What would this look like for mytype? &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'The new way $ Rather than add variant-tests and data-extractors (non-standard jargon), ML has a case expression that uses pattern-matching. In its simplest form, case has one pattern for each constructor in a dataype and binds one variable for each value carried. Example: case e of TwoInts(i1,i2) => e1 | Str s => e2 | Pizza => e3 What are the typing rules? What are the evaluation rules? Patterns are not types nor expressions (despite syntactic similarity) &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Type-checking case $ In addition to binding local variables and requiring branches to have the same type, the typing rules for case prevent some run-time errors: • Exhaustiveness: No test can “fail” (a warning) • Redundancy: No test can be “impossible” (an error) &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Expression trees $ datatype arith_exp = Constant of int | Negate of arith_exp | Add of arith_exp * arith_exp Think of values of type arith_exp as trees where nodes are • Constant with one int child • Negate with one child that can be any arith_exp tree. • Add with two children that can be any arith_exp trees. In general, a type describes a set of values, which are often trees. One-of types give you different variants for nodes. Constructors evaluate arguments to values (trees) and create bigger values (i.e., taller trees). &A Concepts-Focused Introduction to Functional Programming Using Standard ML % 'Where we’re going $ So far, case gives us what we need to use datatypes: • A (combined) way to test variants and extract values • Powerful enough to define our own tests and data-extractors In fact, pattern-matching is far more general and elegant: • Can use it for datatypes already in the top-level environment (e.g., lists and options) • Can use it for any type • Can have deep patterns &A Concepts-Focused Introduction to Functional Programming Using Standard ML %.